Hy²DL is a python library to create hydrological models for rainfall-runoff prediction using deep learning methods. The repository includes implementations with Large-Sample Hydrology datasets such as CAMELS-GB, CAMELS-US, and CAMELS-DE. Besides data-driven architectures, the repository also supports hybrid hydrological models that combine machine learning with process-based knowledge.

The logic of the codes presented here is based on 'NeuralHydrology --- A Python library for Deep Learning research in hydrology' (https://github.com/neuralhydrology/neuralhydrology.git). For a more modular implementation of deep learning method in hydrological modeling we advice the use of Neural Hydrology.

The codes presented in the repository are in the form of python scripts. Additionally several experiments are in the form of JupyterNotebooks for easy reproduction and execution. Following is a quick overview of the repository structure:

• benchmarks: Comparison of our library against other studies from scientific literature.
• data: Folder where the different datasets (e.g CAMELS-GB, CAMELS-US...) should be added. This information should be independently downloaded by the user.
• docs: Library documentation
• examples: Configuration files to run multiple examples.
• notebooks: Jupyter notebooks showing implementation examples, for different cases.
• results: Folder where the results generated by the codes will be stored.
• src/hy2dl: Code of the library.

A release version is available on PyPI and can be installed using:

uv

uv add hy2dl

or pip.

pip install hy2dl

The pyproject.toml file includes the package requirements.

Detailed documentation for the repository can be found at Hy2DL.readthedocs.io.

If you find Hy²DL useful in your research or applications, please cite it as:

@software{acuna2025,
  author       = {Eduardo Acuna and
                  Manuel Álvarez Chaves and
                  Alexander Dolich and
                  Ashish Manoj J},
  title        = {Hy2DL: Hybrid Hydrological modeling using Deep Learning methods},
  year         = 2025,
  publisher    = {Zenodo},
  doi          = {10.5281/zenodo.17251944},
  url          = {https://doi.org/10.5281/zenodo.17251944},
}